{
 "created_utc": "2026-10-18T03:50:49Z",
 "data": {
  "eab_commit": "2961c17ebcb35fd15d5c614c4874d1f567b85d1b",
  "eab_repo": "https://github.com/aisa-group/decomposing-eval-awareness",
  "factorial_sets": "main,factorial,control,ood",
  "geometry_sets": "main,single,leaveout,control,ood",
  "mmlu": {
   "n_items": 600,
   "seed": 0,
   "sha1_all600": "ae8e1e4f8362a0d9b193822d0aa6237aa470a029",
   "sha1_first300": "228ae7b83e7d47af7d786f99c0de68abef7c893a"
  },
  "prompts_sha1_16": "6ba7456f5cb03427",
  "set_counts": {
   "control": 300,
   "factorial": 192,
   "leaveout": 384,
   "main": 200,
   "ood": 30,
   "single": 384
  }
 },
 "files": {
  "KRYTERIA_DO_ZATWIERDZENIA.md": {
   "bytes": 14554,
   "sha256": "b96700593f915dc20a934679bc534ddc6c61f2166057eeeae7c7163e7b0ac3e6"
  },
  "PREREG_QWEN3_32B.md": {
   "bytes": 30408,
   "sha256": "a26ac9bd5935f11045de2eca370a01592801ed4467d872e319c02160e366c860"
  },
  "RUN_CONFIG.json": {
   "bytes": 10994,
   "sha256": "10b88a7cb1101204894187c0b8c793a7c20004886d26280aad1fa68df6528163"
  },
  "bf16_floor_calibration.json": {
   "bytes": 6788,
   "sha256": "52807e65785d48b0b627fd2d9d0a6dfd234ae7bcbd6549ad47eb81802edd484c"
  },
  "calibrate_bf16_floor.py": {
   "bytes": 3789,
   "sha256": "39e92ce88f528e33bc20edf775d07f269c1d2fc6c3987da2572770260f1d334f"
  },
  "colab_q32.py": {
   "bytes": 37066,
   "sha256": "8e95a6687cd267f8401f001ab2936fbc9d24b2e05a30eec80f3961cd63c6894f"
  },
  "compare_mmlu_common.py": {
   "bytes": 3843,
   "sha256": "ebd2bba0bde24ada1daaf43dc3791166d8eda595bd155f5b4561383f5ca9cb51"
  },
  "evaluate_criteria.py": {
   "bytes": 18871,
   "sha256": "88e319cf568b7039bca6ff8ef2746e9c70835e441b9244ef52a6f51bc110980e"
  },
  "freeze_manifest.py": {
   "bytes": 4500,
   "sha256": "d6ed8c49f0e6830e1d463726fb1b7889dc5b315e1b526bbb30ede781b44bc993"
  },
  "memwrap.py": {
   "bytes": 1936,
   "sha256": "0fff12d730bc6799cdc3f6ebb1b49eac0f2fe4c2b027f83a3dcd3155f56c0162"
  },
  "preflight_report.canon.json": {
   "bytes": 16509,
   "sha256": "d4084d2815b1ecd851480f5f9633b53e488189e8299d4bf2d346e06e2dc0a6a8"
  },
  "selftest_interventions.py": {
   "bytes": 7309,
   "sha256": "0b02a56bca7476dc389e20a6f31e847d43355b15b24bf491c0f4012591ada1fc"
  },
  "src/ablation_scan.py": {
   "bytes": 13567,
   "sha256": "fd7066f52755afb6c39adb019f2ed3fbbb50e1bae295375bca14b99f1d1c4cce"
  },
  "src/analyze_mmlu.py": {
   "bytes": 3941,
   "sha256": "c1ff9d515c966e830da71905e8551fd373363791e7b3f6049717453494bcf4ef"
  },
  "src/bootstrap_geometry.py": {
   "bytes": 9977,
   "sha256": "3dec443ce0e1d20bef7980655b1d55f9f69169ea698ad9114f574e211dd774fa"
  },
  "src/build_dataset.py": {
   "bytes": 14408,
   "sha256": "5152daa288ffe1451fbe5c06313239e3afb5f3a0e3500e25f926c8d541c4ea91"
  },
  "src/collect_activations.py": {
   "bytes": 6816,
   "sha256": "199b6fe1beaf9b380667e58bc2dc3832e42a7e28385450d590088efb7f949318"
  },
  "src/compare_directions.py": {
   "bytes": 6468,
   "sha256": "19d0b41c4c4a04d5b4d76fcdaef05d64ace2c7ded8e7a1cc289369b7a87d4d7c"
  },
  "src/crosslayer.py": {
   "bytes": 3256,
   "sha256": "ca4623af78444d079ed8604423dc82763afce34488e6c3f2907c2198a5582b10"
  },
  "src/decorrelate.py": {
   "bytes": 6079,
   "sha256": "ff86e728cb548096a2bc2e6d9eb6cb2adda3b1c0798461d6b662adf7183162a9"
  },
  "src/directions.py": {
   "bytes": 18480,
   "sha256": "a07b860cf59f8fb5adf725c7293fabe1f9e9f12381a645680736fd784161440b"
  },
  "src/judge.py": {
   "bytes": 6554,
   "sha256": "73b962a67dd3949dc68ea2ade6a6c4a8639c8b6145deed0c3571f064471ae4be"
  },
  "src/massive_activations.py": {
   "bytes": 4998,
   "sha256": "341de80803e5cd75260d7c28bd036ab6c8d0b9cb9fc8c93ed0ee41a12901a897"
  },
  "src/mmlu_validate.py": {
   "bytes": 12447,
   "sha256": "cd3bc37a0abe2050dcd93700dfdcb5fc45efa91d7d9f38f9254604c61710c59f"
  },
  "src/model_utils.py": {
   "bytes": 23353,
   "sha256": "61552a3e970325bcc5a56476cb2e6e58427a6cb8253aedc4e633596efb163ef8"
  },
  "src/plots.py": {
   "bytes": 6604,
   "sha256": "e0d0bb9482e618f441bf319c38cbe38a935bfc8de751bd2dae6bc1c822b43a14"
  },
  "src/prepare_mmlu.py": {
   "bytes": 1491,
   "sha256": "2f899875d3d37edccbd70b7ce7cda435e9def79e9b1f8663d6e0af8faaa52915"
  },
  "src/register_factorial.py": {
   "bytes": 8854,
   "sha256": "18f6986ef3b27c9cea264c2eb7c9c11b089e2e538bcdcc7545cf20bee093cdb6"
  },
  "src/steering.py": {
   "bytes": 13116,
   "sha256": "88c4718b32374dae07bf59b6f8108a14f5e1b79ebc08b925fa3b83f4ffefe08c"
  },
  "src/summarize.py": {
   "bytes": 8779,
   "sha256": "18aaa1917f77d9ab4ab9ec938fcddad4a2d4113952d96c45f8d4384a54389a6b"
  },
  "src/test_interventions.py": {
   "bytes": 6414,
   "sha256": "6d023d7393dcf45cf0ca8d9c990d6b22855fcb32816e7066b9869296588fa195"
  },
  "src/verbalize.py": {
   "bytes": 7436,
   "sha256": "fbdb8a0e34de99ee8e2e3c941d05680bbf0ff3987c2a9727fc2d92cf8f911dd6"
  },
  "src/verbalize_grid.py": {
   "bytes": 14415,
   "sha256": "3321589d37814968b3a0668dd896b09c87fd803a6c517edcdf7086226f92a1d5"
  },
  "tools/check_tokenizer.py": {
   "bytes": 4517,
   "sha256": "770d288eee47688fe3ec79104e72cf7707f2d59cad9795624f367e89c457da05"
  },
  "tools/load_probe.py": {
   "bytes": 3505,
   "sha256": "43525ec07d501a56093f316f66b685157961a6f7bd0d954f89290f896f3df349"
  }
 },
 "load": {
  "device_map": {
   "": 0
  },
  "dtype": "bfloat16",
  "env": {
   "EA_DEVICE_MAP_SINGLE": "0"
  },
  "expect_dtype": "bfloat16",
  "forbidden_env": [
   "EA_LOAD_IN_8BIT",
   "EA_LOAD_IN_4BIT",
   "EA_PREQUANTISED",
   "EA_MAX_MEMORY_GB"
  ],
  "local_weights_dir": "/content/qwen3-32b-rev9216db57"
 },
 "manifest_sha256": "b2ec42009c9d4bb1acf3517b0b0d2b5db1c3093848fb402ace97299fae368e10",
 "model": {
  "d_model": 5120,
  "file_bytes": 65524328560,
  "id": "Qwen/Qwen3-32B",
  "n_blocks": 64,
  "n_hidden_states": 65,
  "n_params": 32762123264,
  "n_shards": 17,
  "revision": "9216db5781bf21249d130ec9da846c4624c16137",
  "tensor_bytes": 65524246528,
  "tokenizer_config_sha256": "d5d09f07b48c3086c508b30d1c9114bd1189145b74e982a265350c923acd8101",
  "tokenizer_json_sha256": "aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4"
 },
 "preflight": {
  "chosen": {
   "batch_ablation_scan": 8,
   "batch_collect": 8,
   "batch_mmlu_validate": 8,
   "collect": 8,
   "generation": 8,
   "mmlu_scoring": 16,
   "ppl": 1,
   "vgrid": 1
  },
  "code_sha256": {
   "src/ablation_scan.py": "fd7066f52755afb6c39adb019f2ed3fbbb50e1bae295375bca14b99f1d1c4cce",
   "src/analyze_mmlu.py": "c1ff9d515c966e830da71905e8551fd373363791e7b3f6049717453494bcf4ef",
   "src/bootstrap_geometry.py": "3dec443ce0e1d20bef7980655b1d55f9f69169ea698ad9114f574e211dd774fa",
   "src/build_dataset.py": "5152daa288ffe1451fbe5c06313239e3afb5f3a0e3500e25f926c8d541c4ea91",
   "src/collect_activations.py": "199b6fe1beaf9b380667e58bc2dc3832e42a7e28385450d590088efb7f949318",
   "src/compare_directions.py": "19d0b41c4c4a04d5b4d76fcdaef05d64ace2c7ded8e7a1cc289369b7a87d4d7c",
   "src/crosslayer.py": "ca4623af78444d079ed8604423dc82763afce34488e6c3f2907c2198a5582b10",
   "src/decorrelate.py": "ff86e728cb548096a2bc2e6d9eb6cb2adda3b1c0798461d6b662adf7183162a9",
   "src/directions.py": "a07b860cf59f8fb5adf725c7293fabe1f9e9f12381a645680736fd784161440b",
   "src/judge.py": "73b962a67dd3949dc68ea2ade6a6c4a8639c8b6145deed0c3571f064471ae4be",
   "src/massive_activations.py": "341de80803e5cd75260d7c28bd036ab6c8d0b9cb9fc8c93ed0ee41a12901a897",
   "src/mmlu_validate.py": "cd3bc37a0abe2050dcd93700dfdcb5fc45efa91d7d9f38f9254604c61710c59f",
   "src/model_utils.py": "61552a3e970325bcc5a56476cb2e6e58427a6cb8253aedc4e633596efb163ef8",
   "src/plots.py": "e0d0bb9482e618f441bf319c38cbe38a935bfc8de751bd2dae6bc1c822b43a14",
   "src/prepare_mmlu.py": "2f899875d3d37edccbd70b7ce7cda435e9def79e9b1f8663d6e0af8faaa52915",
   "src/register_factorial.py": "18f6986ef3b27c9cea264c2eb7c9c11b089e2e538bcdcc7545cf20bee093cdb6",
   "src/steering.py": "88c4718b32374dae07bf59b6f8108a14f5e1b79ebc08b925fa3b83f4ffefe08c",
   "src/summarize.py": "18aaa1917f77d9ab4ab9ec938fcddad4a2d4113952d96c45f8d4384a54389a6b",
   "src/test_interventions.py": "6d023d7393dcf45cf0ca8d9c990d6b22855fcb32816e7066b9869296588fa195",
   "src/verbalize.py": "fbdb8a0e34de99ee8e2e3c941d05680bbf0ff3987c2a9727fc2d92cf8f911dd6",
   "src/verbalize_grid.py": "3321589d37814968b3a0668dd896b09c87fd803a6c517edcdf7086226f92a1d5",
   "tools/check_tokenizer.py": "770d288eee47688fe3ec79104e72cf7707f2d59cad9795624f367e89c457da05",
   "tools/load_probe.py": "43525ec07d501a56093f316f66b685157961a6f7bd0d954f89290f896f3df349"
  },
  "env": {
   "cuda": "13.0",
   "disk_free_gib": 193.1,
   "ea_env": {
    "EA_LOAD_IN_4BIT": null,
    "EA_LOAD_IN_8BIT": null,
    "EA_MAX_MEMORY_GB": null,
    "EA_PREQUANTISED": null
   },
   "gpus": [
    {
     "cc": "8.0",
     "gib": 79.25,
     "name": "NVIDIA A100-SXM4-80GB"
    }
   ],
   "platform": "Linux-6.6.122+-x86_64-with-glibc2.39",
   "python": "3.13.15",
   "ram_gib": 167.1,
   "torch": "2.11.0+cu130",
   "transformers": "5.19.0"
  },
  "file": "preflight_report.canon.json",
  "sha256": "d4084d2815b1ecd851480f5f9633b53e488189e8299d4bf2d346e06e2dc0a6a8",
  "stamp": "20261018T013921Z"
 },
 "run": "qwen3-32b-bf16-english-1490",
 "software": {
  "others": "accelerate safetensors numpy scipy scikit-learn pandas matplotlib pyarrow pyyaml huggingface_hub",
  "torch": "as installed on host; recorded in the run manifest",
  "transformers": "==5.19.0"
 },
 "start_confirm_12": "b2ec42009c9d",
 "status": "FROZEN",
 "thresholds_approved": true
}